services/api/app/models/container.rb

   1 # Copyright (C) The Arvados Authors. All rights reserved.
   2 #
   3 # SPDX-License-Identifier: AGPL-3.0
   4
   5 require 'log_reuse_info'
   6 require 'whitelist_update'
   7 require 'safe_json'
   8 require 'update_priority'
   9
  10 class Container < ArvadosModel
  11   include ArvadosModelUpdates
  12   include HasUuid
  13   include KindAndEtag
  14   include CommonApiTemplate
  15   include WhitelistUpdate
  16   extend CurrentApiClient
  17   extend DbCurrentTime
  18   extend LogReuseInfo
  19
  20   # Posgresql JSONB columns should NOT be declared as serialized, Rails 5
  21   # already know how to properly treat them.
  22   attribute :secret_mounts, :jsonbHash, default: {}
  23   attribute :runtime_status, :jsonbHash, default: {}
  24   attribute :runtime_auth_scopes, :jsonbArray, default: []
  25   attribute :output_storage_classes, :jsonbArray, default: lambda { Rails.configuration.DefaultStorageClasses }
  26
  27   serialize :environment, Hash
  28   serialize :mounts, Hash
  29   serialize :runtime_constraints, Hash
  30   serialize :command, Array
  31   serialize :scheduling_parameters, Hash
  32
  33   after_find :fill_container_defaults_after_find
  34   before_validation :fill_field_defaults, :if => :new_record?
  35   before_validation :set_timestamps
  36   before_validation :check_lock
  37   before_validation :check_unlock
  38   validates :command, :container_image, :output_path, :cwd, :priority, { presence: true }
  39   validates :priority, numericality: { only_integer: true, greater_than_or_equal_to: 0 }
  40   validate :validate_runtime_status
  41   validate :validate_state_change
  42   validate :validate_change
  43   validate :validate_lock
  44   validate :validate_output
  45   after_validation :assign_auth
  46   before_save :sort_serialized_attrs
  47   before_save :update_secret_mounts_md5
  48   before_save :scrub_secrets
  49   before_save :clear_runtime_status_when_queued
  50   after_save :update_cr_logs
  51   after_save :handle_completed
  52   after_save :propagate_priority
  53   after_commit { UpdatePriority.run_update_thread }
  54
  55   has_many :container_requests, :foreign_key => :container_uuid, :class_name => 'ContainerRequest', :primary_key => :uuid
  56   belongs_to :auth, :class_name => 'ApiClientAuthorization', :foreign_key => :auth_uuid, :primary_key => :uuid
  57
  58   api_accessible :user, extend: :common do |t|
  59     t.add :command
  60     t.add :container_image
  61     t.add :cwd
  62     t.add :environment
  63     t.add :exit_code
  64     t.add :finished_at
  65     t.add :locked_by_uuid
  66     t.add :log
  67     t.add :mounts
  68     t.add :output
  69     t.add :output_path
  70     t.add :priority
  71     t.add :progress
  72     t.add :runtime_constraints
  73     t.add :runtime_status
  74     t.add :started_at
  75     t.add :state
  76     t.add :auth_uuid
  77     t.add :scheduling_parameters
  78     t.add :runtime_user_uuid
  79     t.add :runtime_auth_scopes
  80     t.add :lock_count
  81     t.add :gateway_address
  82     t.add :interactive_session_started
  83     t.add :output_storage_classes
  84   end
  85
  86   # Supported states for a container
  87   States =
  88     [
  89      (Queued = 'Queued'),
  90      (Locked = 'Locked'),
  91      (Running = 'Running'),
  92      (Complete = 'Complete'),
  93      (Cancelled = 'Cancelled')
  94     ]
  95
  96   State_transitions = {
  97     nil => [Queued],
  98     Queued => [Locked, Cancelled],
  99     Locked => [Queued, Running, Cancelled],
 100     Running => [Complete, Cancelled],
 101     Complete => [Cancelled]
 102   }
 103
 104   def self.limit_index_columns_read
 105     ["mounts"]
 106   end
 107
 108   def self.full_text_searchable_columns
 109     super - ["secret_mounts", "secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 110   end
 111
 112   def self.searchable_columns *args
 113     super - ["secret_mounts_md5", "runtime_token", "gateway_address", "output_storage_classes"]
 114   end
 115
 116   def logged_attributes
 117     super.except('secret_mounts', 'runtime_token')
 118   end
 119
 120   def state_transitions
 121     State_transitions
 122   end
 123
 124   # Container priority is the highest "computed priority" of any
 125   # matching request. The computed priority of a container-submitted
 126   # request is the priority of the submitting container. The computed
 127   # priority of a user-submitted request is a function of
 128   # user-assigned priority and request creation time.
 129   def update_priority!
 130     return if ![Queued, Locked, Running].include?(state)
 131     p = ContainerRequest.
 132         where('container_uuid=? and priority>0', uuid).
 133         includes(:requesting_container).
 134         lock(true).
 135         map do |cr|
 136       if cr.requesting_container
 137         cr.requesting_container.priority
 138       else
 139         (cr.priority << 50) - (cr.created_at.to_time.to_f * 1000).to_i
 140       end
 141     end.max || 0
 142     update_attributes!(priority: p)
 143   end
 144
 145   def propagate_priority
 146     return true unless saved_change_to_priority?
 147     act_as_system_user do
 148       # Update the priority of child container requests to match new
 149       # priority of the parent container (ignoring requests with no
 150       # container assigned, because their priority doesn't matter).
 151       ContainerRequest.
 152         where(requesting_container_uuid: self.uuid,
 153               state: ContainerRequest::Committed).
 154         where('container_uuid is not null').
 155         includes(:container).
 156         map(&:container).
 157         map(&:update_priority!)
 158     end
 159   end
 160
 161   # Create a new container (or find an existing one) to satisfy the
 162   # given container request.
 163   def self.resolve(req)
 164     if req.runtime_token.nil?
 165       runtime_user = if req.modified_by_user_uuid.nil?
 166                        current_user
 167                      else
 168                        User.find_by_uuid(req.modified_by_user_uuid)
 169                      end
 170       runtime_auth_scopes = ["all"]
 171     else
 172       auth = ApiClientAuthorization.validate(token: req.runtime_token)
 173       if auth.nil?
 174         raise ArgumentError.new "Invalid runtime token"
 175       end
 176       runtime_user = User.find_by_id(auth.user_id)
 177       runtime_auth_scopes = auth.scopes
 178     end
 179     c_attrs = act_as_user runtime_user do
 180       {
 181         command: req.command,
 182         cwd: req.cwd,
 183         environment: req.environment,
 184         output_path: req.output_path,
 185         container_image: resolve_container_image(req.container_image),
 186         mounts: resolve_mounts(req.mounts),
 187         runtime_constraints: resolve_runtime_constraints(req.runtime_constraints),
 188         scheduling_parameters: req.scheduling_parameters,
 189         secret_mounts: req.secret_mounts,
 190         runtime_token: req.runtime_token,
 191         runtime_user_uuid: runtime_user.uuid,
 192         runtime_auth_scopes: runtime_auth_scopes,
 193         output_storage_classes: req.output_storage_classes,
 194       }
 195     end
 196     act_as_system_user do
 197       if req.use_existing && (reusable = find_reusable(c_attrs))
 198         reusable
 199       else
 200         Container.create!(c_attrs)
 201       end
 202     end
 203   end
 204
 205   # Return a runtime_constraints hash that complies with requested but
 206   # is suitable for saving in a container record, i.e., has specific
 207   # values instead of ranges.
 208   #
 209   # Doing this as a step separate from other resolutions, like "git
 210   # revision range to commit hash", makes sense only when there is no
 211   # opportunity to reuse an existing container (e.g., container reuse
 212   # is not implemented yet, or we have already found that no existing
 213   # containers are suitable).
 214   def self.resolve_runtime_constraints(runtime_constraints)
 215     rc = {}
 216     runtime_constraints.each do |k, v|
 217       if v.is_a? Array
 218         rc[k] = v[0]
 219       else
 220         rc[k] = v
 221       end
 222     end
 223     if rc['keep_cache_ram'] == 0
 224       rc['keep_cache_ram'] = Rails.configuration.Containers.DefaultKeepCacheRAM
 225     end
 226     rc
 227   end
 228
 229   # Return a mounts hash suitable for a Container, i.e., with every
 230   # readonly collection UUID resolved to a PDH.
 231   def self.resolve_mounts(mounts)
 232     c_mounts = {}
 233     mounts.each do |k, mount|
 234       mount = mount.dup
 235       c_mounts[k] = mount
 236       if mount['kind'] != 'collection'
 237         next
 238       end
 239
 240       uuid = mount.delete 'uuid'
 241
 242       if mount['portable_data_hash'].nil? and !uuid.nil?
 243         # PDH not supplied, try by UUID
 244         c = Collection.
 245           readable_by(current_user).
 246           where(uuid: uuid).
 247           select(:portable_data_hash).
 248           first
 249         if !c
 250           raise ArvadosModel::UnresolvableContainerError.new "cannot mount collection #{uuid.inspect}: not found"
 251         end
 252         mount['portable_data_hash'] = c.portable_data_hash
 253       end
 254     end
 255     return c_mounts
 256   end
 257
 258   # Return a container_image PDH suitable for a Container.
 259   def self.resolve_container_image(container_image)
 260     coll = Collection.for_latest_docker_image(container_image)
 261     if !coll
 262       raise ArvadosModel::UnresolvableContainerError.new "docker image #{container_image.inspect} not found"
 263     end
 264     coll.portable_data_hash
 265   end
 266
 267   def self.find_reusable(attrs)
 268     log_reuse_info { "starting with #{Container.all.count} container records in database" }
 269     candidates = Container.where_serialized(:command, attrs[:command], md5: true)
 270     log_reuse_info(candidates) { "after filtering on command #{attrs[:command].inspect}" }
 271
 272     candidates = candidates.where('cwd = ?', attrs[:cwd])
 273     log_reuse_info(candidates) { "after filtering on cwd #{attrs[:cwd].inspect}" }
 274
 275     candidates = candidates.where_serialized(:environment, attrs[:environment], md5: true)
 276     log_reuse_info(candidates) { "after filtering on environment #{attrs[:environment].inspect}" }
 277
 278     candidates = candidates.where('output_path = ?', attrs[:output_path])
 279     log_reuse_info(candidates) { "after filtering on output_path #{attrs[:output_path].inspect}" }
 280
 281     image = resolve_container_image(attrs[:container_image])
 282     candidates = candidates.where('container_image = ?', image)
 283     log_reuse_info(candidates) { "after filtering on container_image #{image.inspect} (resolved from #{attrs[:container_image].inspect})" }
 284
 285     candidates = candidates.where_serialized(:mounts, resolve_mounts(attrs[:mounts]), md5: true)
 286     log_reuse_info(candidates) { "after filtering on mounts #{attrs[:mounts].inspect}" }
 287
 288     secret_mounts_md5 = Digest::MD5.hexdigest(SafeJSON.dump(self.deep_sort_hash(attrs[:secret_mounts])))
 289     candidates = candidates.where('secret_mounts_md5 = ?', secret_mounts_md5)
 290     log_reuse_info(candidates) { "after filtering on secret_mounts_md5 #{secret_mounts_md5.inspect}" }
 291
 292     if attrs[:runtime_constraints]['cuda'].nil?
 293       attrs[:runtime_constraints]['cuda'] = {
 294         'device_count' => 0,
 295         'driver_version' => '',
 296         'hardware_capability' => '',
 297       }
 298     end
 299     resolved_runtime_constraints = [resolve_runtime_constraints(attrs[:runtime_constraints])]
 300     if resolved_runtime_constraints[0]['cuda']['device_count'] == 0
 301       # If no CUDA requested, extend search to include older container
 302       # records that don't have a 'cuda' section in runtime_constraints
 303       resolved_runtime_constraints << resolved_runtime_constraints[0].except('cuda')
 304     end
 305
 306     candidates = candidates.where_serialized(:runtime_constraints, resolved_runtime_constraints, md5: true, multivalue: true)
 307     log_reuse_info(candidates) { "after filtering on runtime_constraints #{attrs[:runtime_constraints].inspect}" }
 308
 309     log_reuse_info { "checking for state=Complete with readable output and log..." }
 310
 311     select_readable_pdh = Collection.
 312       readable_by(current_user).
 313       select(:portable_data_hash).
 314       to_sql
 315
 316     usable = candidates.where(state: Complete, exit_code: 0)
 317     log_reuse_info(usable) { "with state=Complete, exit_code=0" }
 318
 319     usable = usable.where("log IN (#{select_readable_pdh})")
 320     log_reuse_info(usable) { "with readable log" }
 321
 322     usable = usable.where("output IN (#{select_readable_pdh})")
 323     log_reuse_info(usable) { "with readable output" }
 324
 325     usable = usable.order('finished_at ASC').limit(1).first
 326     if usable
 327       log_reuse_info { "done, reusing container #{usable.uuid} with state=Complete" }
 328       return usable
 329     end
 330
 331     # Check for non-failing Running candidates and return the most likely to finish sooner.
 332     log_reuse_info { "checking for state=Running..." }
 333     running = candidates.where(state: Running).
 334               where("(runtime_status->'error') is null").
 335               order('progress desc, started_at asc').
 336               limit(1).first
 337     if running
 338       log_reuse_info { "done, reusing container #{running.uuid} with state=Running" }
 339       return running
 340     else
 341       log_reuse_info { "have no containers in Running state" }
 342     end
 343
 344     # Check for Locked or Queued ones and return the most likely to start first.
 345     locked_or_queued = candidates.
 346                        where("state IN (?)", [Locked, Queued]).
 347                        order('state asc, priority desc, created_at asc').
 348                        limit(1).first
 349     if locked_or_queued
 350       log_reuse_info { "done, reusing container #{locked_or_queued.uuid} with state=#{locked_or_queued.state}" }
 351       return locked_or_queued
 352     else
 353       log_reuse_info { "have no containers in Locked or Queued state" }
 354     end
 355
 356     log_reuse_info { "done, no reusable container found" }
 357     nil
 358   end
 359
 360   def lock
 361     self.with_lock do
 362       if self.state != Queued
 363         raise LockFailedError.new("cannot lock when #{self.state}")
 364       end
 365       self.update_attributes!(state: Locked)
 366     end
 367   end
 368
 369   def check_lock
 370     if state_was == Queued and state == Locked
 371       if self.priority <= 0
 372         raise LockFailedError.new("cannot lock when priority<=0")
 373       end
 374       self.lock_count = self.lock_count+1
 375     end
 376   end
 377
 378   def unlock
 379     self.with_lock do
 380       if self.state != Locked
 381         raise InvalidStateTransitionError.new("cannot unlock when #{self.state}")
 382       end
 383       self.update_attributes!(state: Queued)
 384     end
 385   end
 386
 387   def check_unlock
 388     if state_was == Locked and state == Queued
 389       if self.locked_by_uuid != current_api_client_authorization.uuid
 390         raise ArvadosModel::PermissionDeniedError.new("locked by a different token")
 391       end
 392       if self.lock_count >= Rails.configuration.Containers.MaxDispatchAttempts
 393         self.state = Cancelled
 394         self.runtime_status = {error: "Failed to start container.  Cancelled after exceeding 'Containers.MaxDispatchAttempts' (lock_count=#{self.lock_count})"}
 395       end
 396     end
 397   end
 398
 399   def self.readable_by(*users_list)
 400     # Load optional keyword arguments, if they exist.
 401     if users_list.last.is_a? Hash
 402       kwargs = users_list.pop
 403     else
 404       kwargs = {}
 405     end
 406     if users_list.select { |u| u.is_admin }.any?
 407       return super
 408     end
 409     Container.where(ContainerRequest.readable_by(*users_list).where("containers.uuid = container_requests.container_uuid").arel.exists)
 410   end
 411
 412   def final?
 413     [Complete, Cancelled].include?(self.state)
 414   end
 415
 416   def self.for_current_token
 417     return if !current_api_client_authorization
 418     _, _, _, container_uuid = Thread.current[:token].split('/')
 419     if container_uuid.nil?
 420       Container.where(auth_uuid: current_api_client_authorization.uuid).first
 421     else
 422       Container.where('auth_uuid=? or (uuid=? and runtime_token=?)',
 423                       current_api_client_authorization.uuid,
 424                       container_uuid,
 425                       current_api_client_authorization.token).first
 426     end
 427   end
 428
 429   protected
 430
 431   def fill_field_defaults
 432     self.state ||= Queued
 433     self.environment ||= {}
 434     self.runtime_constraints ||= {}
 435     self.mounts ||= {}
 436     self.cwd ||= "."
 437     self.priority ||= 0
 438     self.scheduling_parameters ||= {}
 439   end
 440
 441   def permission_to_create
 442     current_user.andand.is_admin
 443   end
 444
 445   def permission_to_destroy
 446     current_user.andand.is_admin
 447   end
 448
 449   def ensure_owner_uuid_is_permitted
 450     # validate_change ensures owner_uuid can't be changed at all --
 451     # except during create, which requires admin privileges. Checking
 452     # permission here would be superfluous.
 453     true
 454   end
 455
 456   def set_timestamps
 457     if self.state_changed? and self.state == Running
 458       self.started_at ||= db_current_time
 459     end
 460
 461     if self.state_changed? and [Complete, Cancelled].include? self.state
 462       self.finished_at ||= db_current_time
 463     end
 464   end
 465
 466   # Check that well-known runtime status keys have desired data types
 467   def validate_runtime_status
 468     [
 469       'error', 'errorDetail', 'warning', 'warningDetail', 'activity'
 470     ].each do |k|
 471       if self.runtime_status.andand.include?(k) && !self.runtime_status[k].is_a?(String)
 472         errors.add(:runtime_status, "'#{k}' value must be a string")
 473       end
 474     end
 475   end
 476
 477   def validate_change
 478     permitted = [:state]
 479     progress_attrs = [:progress, :runtime_status, :log, :output]
 480     final_attrs = [:exit_code, :finished_at]
 481
 482     if self.new_record?
 483       permitted.push(:owner_uuid, :command, :container_image, :cwd,
 484                      :environment, :mounts, :output_path, :priority,
 485                      :runtime_constraints, :scheduling_parameters,
 486                      :secret_mounts, :runtime_token,
 487                      :runtime_user_uuid, :runtime_auth_scopes,
 488                      :output_storage_classes)
 489     end
 490
 491     case self.state
 492     when Locked
 493       permitted.push :priority, :runtime_status, :log, :lock_count
 494
 495     when Queued
 496       permitted.push :priority
 497
 498     when Running
 499       permitted.push :priority, *progress_attrs
 500       if self.state_changed?
 501         permitted.push :started_at, :gateway_address
 502       end
 503       if !self.interactive_session_started_was
 504         permitted.push :interactive_session_started
 505       end
 506
 507     when Complete
 508       if self.state_was == Running
 509         permitted.push *final_attrs, *progress_attrs
 510       end
 511
 512     when Cancelled
 513       case self.state_was
 514       when Running
 515         permitted.push :finished_at, *progress_attrs
 516       when Queued, Locked
 517         permitted.push :finished_at, :log, :runtime_status
 518       end
 519
 520     else
 521       # The state_transitions check will add an error message for this
 522       return false
 523     end
 524
 525     if self.state_was == Running &&
 526        !current_api_client_authorization.nil? &&
 527        (current_api_client_authorization.uuid == self.auth_uuid ||
 528         current_api_client_authorization.token == self.runtime_token)
 529       # The contained process itself can write final attrs but can't
 530       # change priority or log.
 531       permitted.push *final_attrs
 532       permitted = permitted - [:log, :priority]
 533     elsif !current_user.andand.is_admin
 534       raise PermissionDeniedError
 535     elsif self.locked_by_uuid && self.locked_by_uuid != current_api_client_authorization.andand.uuid
 536       # When locked, progress fields cannot be updated by the wrong
 537       # dispatcher, even though it has admin privileges.
 538       permitted = permitted - progress_attrs
 539     end
 540     check_update_whitelist permitted
 541   end
 542
 543   def validate_lock
 544     if [Locked, Running].include? self.state
 545       # If the Container was already locked, locked_by_uuid must not
 546       # changes. Otherwise, the current auth gets the lock.
 547       need_lock = locked_by_uuid_was || current_api_client_authorization.andand.uuid
 548     else
 549       need_lock = nil
 550     end
 551
 552     # The caller can provide a new value for locked_by_uuid, but only
 553     # if it's exactly what we expect. This allows a caller to perform
 554     # an update like {"state":"Unlocked","locked_by_uuid":null}.
 555     if self.locked_by_uuid_changed?
 556       if self.locked_by_uuid != need_lock
 557         return errors.add :locked_by_uuid, "can only change to #{need_lock}"
 558       end
 559     end
 560     self.locked_by_uuid = need_lock
 561   end
 562
 563   def validate_output
 564     # Output must exist and be readable by the current user.  This is so
 565     # that a container cannot "claim" a collection that it doesn't otherwise
 566     # have access to just by setting the output field to the collection PDH.
 567     if output_changed?
 568       c = Collection.
 569             readable_by(current_user, {include_trash: true}).
 570             where(portable_data_hash: self.output).
 571             first
 572       if !c
 573         errors.add :output, "collection must exist and be readable by current user."
 574       end
 575     end
 576   end
 577
 578   def update_cr_logs
 579     # If self.final?, this update is superfluous: the final log/output
 580     # update will be done when handle_completed calls finalize! on
 581     # each requesting CR.
 582     return if self.final? || !saved_change_to_log?
 583     leave_modified_by_user_alone do
 584       ContainerRequest.where(container_uuid: self.uuid).each do |cr|
 585         cr.update_collections(container: self, collections: ['log'])
 586         cr.save!
 587       end
 588     end
 589   end
 590
 591   def assign_auth
 592     if self.auth_uuid_changed?
 593          return errors.add :auth_uuid, 'is readonly'
 594     end
 595     if not [Locked, Running].include? self.state
 596       # Don't need one. If auth already exists, expire it.
 597       #
 598       # We use db_transaction_time here (not db_current_time) to
 599       # ensure the token doesn't validate later in the same
 600       # transaction (e.g., in a test case) by satisfying expires_at >
 601       # transaction timestamp.
 602       self.auth.andand.update_attributes(expires_at: db_transaction_time)
 603       self.auth = nil
 604       return
 605     elsif self.auth
 606       # already have one
 607       return
 608     end
 609     if self.runtime_token.nil?
 610       if self.runtime_user_uuid.nil?
 611         # legacy behavior, we don't have a runtime_user_uuid so get
 612         # the user from the highest priority container request, needed
 613         # when performing an upgrade and there are queued containers,
 614         # and some tests.
 615         cr = ContainerRequest.
 616                where('container_uuid=? and priority>0', self.uuid).
 617                order('priority desc').
 618                first
 619         if !cr
 620           return errors.add :auth_uuid, "cannot be assigned because priority <= 0"
 621         end
 622         self.runtime_user_uuid = cr.modified_by_user_uuid
 623         self.runtime_auth_scopes = ["all"]
 624       end
 625
 626       # Generate a new token. This runs with admin credentials as it's done by a
 627       # dispatcher user, so expires_at isn't enforced by API.MaxTokenLifetime.
 628       self.auth = ApiClientAuthorization.
 629                     create!(user_id: User.find_by_uuid(self.runtime_user_uuid).id,
 630                             api_client_id: 0,
 631                             scopes: self.runtime_auth_scopes)
 632     end
 633   end
 634
 635   def sort_serialized_attrs
 636     if self.environment_changed?
 637       self.environment = self.class.deep_sort_hash(self.environment)
 638     end
 639     if self.mounts_changed?
 640       self.mounts = self.class.deep_sort_hash(self.mounts)
 641     end
 642     if self.runtime_constraints_changed?
 643       self.runtime_constraints = self.class.deep_sort_hash(self.runtime_constraints)
 644     end
 645     if self.scheduling_parameters_changed?
 646       self.scheduling_parameters = self.class.deep_sort_hash(self.scheduling_parameters)
 647     end
 648     if self.runtime_auth_scopes_changed?
 649       self.runtime_auth_scopes = self.runtime_auth_scopes.sort
 650     end
 651   end
 652
 653   def update_secret_mounts_md5
 654     if self.secret_mounts_changed?
 655       self.secret_mounts_md5 = Digest::MD5.hexdigest(
 656         SafeJSON.dump(self.class.deep_sort_hash(self.secret_mounts)))
 657     end
 658   end
 659
 660   def scrub_secrets
 661     # this runs after update_secret_mounts_md5, so the
 662     # secret_mounts_md5 will still reflect the secrets that are being
 663     # scrubbed here.
 664     if self.state_changed? && self.final?
 665       self.secret_mounts = {}
 666       self.runtime_token = nil
 667     end
 668   end
 669
 670   def clear_runtime_status_when_queued
 671     # Avoid leaking status messages between different dispatch attempts
 672     if self.state_was == Locked && self.state == Queued
 673       self.runtime_status = {}
 674     end
 675   end
 676
 677   def handle_completed
 678     # This container is finished so finalize any associated container requests
 679     # that are associated with this container.
 680     if saved_change_to_state? and self.final?
 681       # These get wiped out by with_lock (which reloads the record),
 682       # so record them now in case we need to schedule a retry.
 683       prev_secret_mounts = secret_mounts_before_last_save
 684       prev_runtime_token = runtime_token_before_last_save
 685
 686       # Need to take a lock on the container to ensure that any
 687       # concurrent container requests that might try to reuse this
 688       # container will block until the container completion
 689       # transaction finishes.  This ensure that concurrent container
 690       # requests that try to reuse this container are finalized (on
 691       # Complete) or don't reuse it (on Cancelled).
 692       self.with_lock do
 693         act_as_system_user do
 694           if self.state == Cancelled
 695             retryable_requests = ContainerRequest.where("container_uuid = ? and priority > 0 and state = 'Committed' and container_count < container_count_max", uuid)
 696           else
 697             retryable_requests = []
 698           end
 699
 700           if retryable_requests.any?
 701             c_attrs = {
 702               command: self.command,
 703               cwd: self.cwd,
 704               environment: self.environment,
 705               output_path: self.output_path,
 706               container_image: self.container_image,
 707               mounts: self.mounts,
 708               runtime_constraints: self.runtime_constraints,
 709               scheduling_parameters: self.scheduling_parameters,
 710               secret_mounts: prev_secret_mounts,
 711               runtime_token: prev_runtime_token,
 712               runtime_user_uuid: self.runtime_user_uuid,
 713               runtime_auth_scopes: self.runtime_auth_scopes
 714             }
 715             c = Container.create! c_attrs
 716             retryable_requests.each do |cr|
 717               cr.with_lock do
 718                 leave_modified_by_user_alone do
 719                   # Use row locking because this increments container_count
 720                   cr.container_uuid = c.uuid
 721                   cr.save!
 722                 end
 723               end
 724             end
 725           end
 726
 727           # Notify container requests associated with this container
 728           ContainerRequest.where(container_uuid: uuid,
 729                                  state: ContainerRequest::Committed).each do |cr|
 730             leave_modified_by_user_alone do
 731               cr.finalize!
 732             end
 733           end
 734
 735           # Cancel outstanding container requests made by this container.
 736           ContainerRequest.
 737             includes(:container).
 738             where(requesting_container_uuid: uuid,
 739                   state: ContainerRequest::Committed).each do |cr|
 740             leave_modified_by_user_alone do
 741               cr.update_attributes!(priority: 0)
 742               cr.container.reload
 743               if cr.container.state == Container::Queued || cr.container.state == Container::Locked
 744                 # If the child container hasn't started yet, finalize the
 745                 # child CR now instead of leaving it "on hold", i.e.,
 746                 # Queued with priority 0.  (OTOH, if the child is already
 747                 # running, leave it alone so it can get cancelled the
 748                 # usual way, get a copy of the log collection, etc.)
 749                 cr.update_attributes!(state: ContainerRequest::Final)
 750               end
 751             end
 752           end
 753         end
 754       end
 755     end
 756   end
 757 end